Channel assembling and resource allocation in multichannel spectrum sharing wireless networks

(1)

U

NIVERSITY OF THE

W

ITWATERSRAND

Channel Assembling and Resource Allocation

in Multichannel Spectrum Sharing

Wireless Networks

Chabalala Stephen Chabalala

(2)

C

HANNEL

A

SSEMBLING AND

R

ESOURCE

A

LLOCATION

IN

M

ULTICHANNEL

S

PECTRUM

S

HARING

W

IRELESS

N

ETWORKS

Chabalala Stephen

C

HABALALA

Supervised by

Professor Fambirai

T

AKAWIRA

Co-supervised by

Professor Rex

VAN

O

LST

Submitted in fulfilment of the academic requirements for the degree of Doctor of Philosophy (Ph.D.) in Engineering, in the School of Electrical and Information Engineering, Faculty of Engineering and the Built Environment,

at the University of the Witwatersrand, Johannesburg, SOUTH AFRICA.

(3)

CHABALALA S.C. i 2017 Ph.D. THESIS

A

UTHORIZATION

As the candidate’s supervisor, I have approved this thesis for submission. Name: Professor Fambirai TAKAWIRA

Signed: _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ Date: _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

As the candidate’s co-supervisor, I have approved this thesis for submission. Name: Professor Rex VAN OLST

Signed: _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ Date: _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

(4)

CHABALALA S.C. ii 2017 Ph.D. THESIS

DECLARATION

I, CHABALALA STEPHEN CHABALALA declare that

i. The research reported in this thesis, except where otherwise indicated, is my original work. ii. This thesis has not been submitted for any degree or examination at any other university. iii. This thesis does not contain other persons’ data, pictures, graphs or other information, unless

specifically acknowledged as being sourced from other persons.

iv. This thesis does not contain other persons’ writing, unless specifically acknowledged as being sourced from other researchers. Where other written sources have been quoted, then: a. their words have been re-written but the general information attributed to them has been

referenced;

b. where their exact words have been used, their writing has been placed inside quotation marks, and referenced.

v. Where I have reproduced a publication of which I am an author, co-author or editor, I have indicated in detail which part of the publication was actually written by myself alone and have fully referenced such publications.

vi. This thesis does not contain text, graphics or tables copied and pasted from the Internet, unless specifically acknowledged, and the source being detailed in the thesis and in the ‘List of References’ sections.

Signed:

Date: Wednesday 07, June 2017

(5)

CHABALALA S.C. iii 2017 Ph.D. THESIS

D

EDICATION

(6)

CHABALALA S.C. iv 2017 Ph.D. THESIS

P

REFACE

THIS thesis presents my Ph.D. research study on dynamic resource allocation and channel assembling techniques in spectrum sharing wireless networks. The work has been done under the joint supervision of Professor Fambirai Takawira and Professor Rex van Olst, in the School of Electrical and Information Engineering, at the University of the Witwatersrand (Wits University), Johannesburg, South Africa. This work was supported in part, by the Center for Telecommunications Access and Services (CeTAS) at Wits University.

In a nutshell, this work is focused on the development of dynamic and adaptive resource allocation and management techniques for spectrum sharing and the emerging wireless networks. Accordingly, this research has resulted in the new concepts and techniques on how to perform effective spectrum characterization with respect to time varying nature of wireless channels and the activity patterns of licensed users; how to perform power control and interference management techniques for efficient use of the limited radio spectrum resource; as well as establishing mathematical models for performance analysis over fading wireless channels.

Chabalala S. CHABALALA

University of the Witwatersrand Johannesburg, Wits 2050, South Africa. Wednesday 07, June 2017

(7)

CHABALALA S.C. v 2017 Ph.D. THESIS

ACKNOWLEDGEMENTS

THE work presented in this thesis would not have been accomplished if it were not because of the generous support and the assistance I received from many people, who throughout the challenging years of my Ph.D. study, have been kind. To all, I am very grateful.

First of all, I would like to express my sincere gratitude and appreciation to my principal supervisor, Professor Fambirai Takawira, who has always been the anchor, and offered me the incredible support throughout the entire ordeal. Without the requisite assistance I received from him, the work presented herein would not have seen the light of the day. I would also like to thank my co-supervisor, Professor Rex van Olst for his insightful guidance and support, which I received even beyond his retirement during my research study.

I would like to extend my gratitude to the entire academic staff, the administrative staff and the technical staff members in the School of Electrical and Information Engineering (EIE) for all their support and assistance. Many thanks to my fellows in the Center for Telecommunications Access and Services (CeTAS) research group for the technical discussions and great collaboration.

I would also like to thank my proposal defence committee, for the initial approval of my Ph.D. research study. In particular, thanks to Dr. Olutayo Oyerinde for organising my proposal defence presentation, and Prof. Ekow Otoo for consolidating feedback and comments from the review panel. To Prof. Ivan Hofsajer, the ‘MVP’ (i.e. minimum viable project) approach has been helpful, thank you for your advice. To Jacques Naude, thank you for the insightful discussions on probability theory. Many thanks to Prof. Estelle Trengove for her recommendation, encouragement and assistance in teaching whilst conducting my research study. Great appreciation also goes to the anonymous examiners for their meticulous review and insightful comments.

Special thanks to my wonderful wife ’Matabane Chabalala, and amazing son, Tabane Chabalala, for their outstanding commitment to make this Ph.D. study a success. My heartfelt gratitude goes to my father who could not see this thesis completed, Ntate Matšerane Chabalala, and my caring mother, ’M`e ’Mankalimeng Chabalala, for her ongoing love and endless support. To my siblings, Thato, ’Mantho and Refiloe, I thank you all for your great support. Kea leboha.

(8)

CHABALALA S.C. vi 2017 Ph.D. THESIS

ABSTRACT

HE continuous evolution of wireless communications technologies has increasingly imposed a burden on the use of radio spectrum. Due to the proliferation of new wireless networks applications and services, the radio spectrum is getting saturated and becoming a limited resource. To a large extent, spectrum scarcity may be a result of deficient spectrum allocation and management policies, rather than of the physical shortage of radio frequencies. The conventional static spectrum allocation has been found to be ineffective, leading to overcrowding and inefficient use. Cognitive radio (CR) has therefore emerged as an enabling technology that facilitates dynamic spectrum access (DSA), with a great potential to address the issue of spectrum scarcity and inefficient use. However, provisioning of reliable and robust communication with seamless operation in cognitive radio networks (CRNs) is a challenging task. The underlying challenges include development of non-intrusive dynamic resource allocation (DRA) and optimization techniques.

The main focus of this thesis is development of adaptive channel assembling (ChA) and DRA schemes, with the aim to maximize performance of secondary user (SU) nodes in CRNs, without degrading performance of primary user (PU) nodes in a primary network (PN). The key objectives are therefore four-fold. Firstly, to optimize ChA and DRA schemes in overlay CRNs. Secondly, to develop analytical models for quantifying performance of ChA schemes over fading channels in overlay CRNs. Thirdly, to extend the overlay ChA schemes into hybrid overlay and underlay architectures, subject to power control and interference mitigation; and finally, to extend the adaptive ChA and DRA schemes for multiuser multichannel access CRNs.

Performance analysis and evaluation of the developed ChA and DRA is presented, mainly through extensive simulations and analytical models. Further, the cross validation has been performed between simulations and analytical results to confirm the accuracy and preciseness of the novel analytical models developed in this thesis. In general, the presented results demonstrate improved performance of SU nodes in terms of capacity, collision probability, outage probability and forced termination probability when employing the adaptive ChA and DRA in CRNs.

Index-Terms___Channel assembling (ChA); cognitive radio networks (CRNs); Dynamic spectrum access (DSA); multiuser multichannel access; radio resource management (RRM).

(9)

CHABALALA S.C. vii 2017 Ph.D. THESIS

LIST OF FIGURES

Figure 1-1: Basic cognitive radio functions and operational cycle [13]. ... 2

Figure 1-2: Resource allocation and optimization process in CRNs. ... 3

Figure 1-3: Illustration of channel-bonding and channel-aggregation. ... 5

Figure 2-1: The concept of spectrum holes in CRNs [13]. ... 12

Figure 2-2: Example of underlay spectrum access in CRNs. ... 14

Figure 2-3: Example of hybrid spectrum access in CRNs. ... 15

Figure 2-4: Illustration of the impact of power allocation on ChA techniques. ... 18

Figure 3-1: Network model for coexistence of a PN and CRN. ... 25

Figure 3-2: Collision probability between PU and SU services for increasing PU arrival rates... 30

Figure 3-3: Illustration of feasibility-region for optimal solution. ... 32

Figure 3-4: SU average capacity for increasing PU arrival rates at = 0.9. ... 43

Figure 3-5: SU average capacity for increasing PU arrival rates at = 0.8. ... 43

Figure 3-6: SU outage probability for increasing PU arrival rates at = 0.9... 44

Figure 3-7: SU outage probability for increasing PU arrival rates at = 0.8... 44

Figure 3-8: SU collision probability for increasing PU arrival rates at = 0.9. ... 45

Figure 3-9: SU collision probability for increasing PU arrival rates at = 0.8. ... 45

Figure 4-1: SU average capacity for increasing PU arrival rates. ... 59

(15)

CHABALALA S.C. xiii 2017 Ph.D. THESIS

Figure 4-3: Forced-termination probability for increasing PU arrivals. ... 60

Figure 4-4: Average capacity against transmit power. ... 60

Figure 4-5: SU outage probability against transmit power. ... 61

Figure 4-6: Forced-termination against transmit power. ... 61

Figure 4-7: Outage probability against capacity threshold. ... 62

Figure 5-1: Network model for hybrid overlay and underlay SU transmissions. ... 65

Figure 5-2: Channel capacity for increasing PU arrival rates. ... 82

Figure 5-3: SU Outage probability for increasing PU arrival rates. ... 82

Figure 5-4: Collision probability for increasing PU arrival rates. ... 83

Figure 5-5: SU channel capacity versus SU total transmit power. ... 83

Figure 5-6: Channel capacity for increasing PU arrival rates. ... 84

Figure 5-7: Collision probability versus SU total transmit power. ... 84

Figure 5-8: Outage probability versus SU QoS requirement on capacity. ... 85

Figure 6-1: Network model for infrastructure based multiuser multichannel CRN. ... 88

Figure 6-2: Average SU capacity vs PU arrivals. ... 99

Figure 6-3: SU outage probability vs PU arrivals. ... 99

Figure 6-4: SU collision probability vs PU arrivals. ... 100

Figure 6-5: SU average capacity vs total transmit power. ... 100

Figure 6-6: SU outage probability vs total transmit power. ... 101

(16)

CHABALALA S.C. xiv 2017 Ph.D. THESIS

LIST OF TABLES

Table 3-1: Computational Complexity for Single-SU ChA Schemes ... 40

Table 3-2: Simulation Parameters for Overlay Spectrum Access ... 41

Table 5-1: Hybrid Overlay and Underlay ChA Schemes ... 80

Table 5-2: Hybrid Overlay and Underlay ChA Schemes ... 80

Table 6-1: Computational Complexity for Multi-SU ChA Schemes ... 96

Table 6-2: Multiuser Access ChA Schemes ... 97

(17)

CHABALALA S.C. xv 2017 Ph.D. THESIS

LIST OF ACRONYMS

AWGN: Additive White Gaussian Noise BnB: Branch-and-Bound

BnBKOP: Branch-and-Bound-K Optimal Power ChA: Channel Assembling

CDF: Cumulative Distribution Function

CR: Cognitive Radio

CRN: Cognitive Radio Network CSA: Concurrent Spectrum Access CSI: Channel Status Information CTMC: Continuous Time Markov Chain DRA: Dynamic Resource Allocation DSA: Dynamic Spectrum Access

FCC: Federal Communications Commission FixedKEP: Fixed-K Equal Power

FixedKOP: Fixed-K Optimal Power HunKOP: Hungarian-K Optimal Power

IEEE: Institute of Electrical and Electronics Engineering IET: Institute of Engineering and Technology

i.i.d.: Independent and Identically Distributed

i.n.i.d: Independent Non-Identically Distributed KKT: Karush-Kuhn-Tucker

LB: Lower Bound

(18)

CHABALALA S.C. xvi 2017 Ph.D. THESIS

MATLAB: Matrix Laboratory

OFDM: Orthogonal Frequency Division Multiplexing OSA: Opportunistic Spectrum Access

PDF: Probability Distribution Function PGF: Probability Generating Function

PN: Primary Network

POMDP: Partially Observable Markov Decision Process

PU: Primary User

QoS: Quality of Service

RndKOP: Rounded-K Optimal Power RRM: Radio Resource Management SF: Sequential-Fixing

SNIR: Signal-noise and Interference Ratio SNR: Signal-to-Noise Ratio

SN: Secondary Network

SU: Secondary User

UB: Upper Bound

(19)

CHABALALA S.C. xvii 2017 Ph.D. THESIS

LIST OF SYMBOLS

ℬ Bandwidth

Set of total PU channels Set of occupied PU channels

Number of busy PU channels Set of free PU channels Number of free PU channels Channel power gain vector Number of assembled channels Channel selection vector Power vector

Set of SU nodes

Number of SU nodes in

( , ) Capacity over channel

Gaussian background noise

! SNR on channel

"#∙% Expectation operator

&(∙) Unit-step function

ℛ ( , ) Sum capacity of assembled channels

∆) Number of function evaluations ∆* Number of iterations

+ Iteration index

, Set of assembled channels

∆- Time interval

(20)

CHABALALA S.C. xviii 2017 Ph.D. THESIS

./0 Transmission time

12343 Data length 567 PU arrival rate

8 Collision probability threshold 9 Collision probability characterization ℛ8 Minimum capacity threshold

:;30 Maximum available transmit power

max#?, @% Maximum of ? and @

AB#∙% Mellin transform

Г(D, E) Complementary incomplete gamma function

ℋF,GH,IJ∙K Fox-ℋ function

L(∙) Lambert-L function

F,GH,IJ∙K Meijer- function

ℙN#∙% Probability operator

.O,67 PU arrival time

PQ(∙) Probability generating function

:67 PU transmit power

R67S PU noise variance RB7S SU noise variance !67 SNR at the PU receiver

!TU SNR threshold

;30,Q SU maximum allowable transmit power VTU Outage probability threshold

(21)

CHABALALA S.C. 1 2017 Ph.D. THESIS

C

HAPTER

O

NE

1

I

NTRODUCTION

1.1

General Background

THE rapid advances in both hardware and software technologies fostered the proliferation of wireless networks devices, as well as fueling the exponential growth of new applications and services [1]-[2]. The next-generation wireless networks and the emerging technologies are envisioned to provide reliable and robust communication with seamless operation. Further, the emergence of bandwidth-hungry network applications and services with varying quality-of-service (QoS) requirements has further resulted in explosive demands for ubiquitous high-speed wireless networks, which underline the need for new communication techniques that are capable of transporting large amounts of data with varying QoS requirements [1], [3]-[4].

In recent research efforts, accessibility and reliability of communication services have been established as crucial aspects for future wireless networks [5]-[6]. However, the proliferation of wireless networks devices poses significant challenges that lead to scarcity of radio spectrum, together with inefficient utilization due to the traditional static spectrum management policies. To a large extent, spectrum scarcity is due to deficient spectrum allocation and management policies, rather than of the physical shortage of radio frequencies [2], [5] [7].

Efficient radio resource management (RRM) techniques are mandatory for provisioning of robust and ultra-reliable communication (URC). In essence, RRM techniques are required to maintain high system utilization while satisfying users QoS requirements [8]-[11]. URC is one of the challenging concepts for which the emerging wireless networks are envisaged to address. In principle, URC concept refers to the provisioning of communication services at the certain QoS level with a high degree of dependability, whereby dependability is attributed to high service availability and reliability [7]-[8], [11]. Consequently, the main focus of this research is to study and quantify performance of, as well as developing new concepts and techniques for adaptive channel assembling (ChA) and resource allocation schemes in spectrum sharing wireless networks.

(22)

CHABALALA S.C. 2 2017 Ph.D. THESIS 1.1.1 Cognitive Radio Networks

Cognitive radio (CR) constitutes an inventive technology that enables intelligent operation in wireless networks, whereby a CR node can adapt its operating parameters based on environmental conditions and network constraints as shown in Fig. 1-1 [11]-[12]. The two main components that are significant for successful operation of cognitive radio networks (CRNs) are spectrum sensing, which detects spectrum holes in a primary network (PN); and spectrum allocation, which deals with how to allocate spectrum holes to secondary user (SU) nodes [1], [7]. Spectrum holes are the unoccupied bands of frequencies that are assigned to primary user (PU) nodes.

Through CR concept, dynamic spectrum access (DSA) allows seamless coexistence of PU nodes and SU nodes sharing radio spectrum. The coexistence offers a great potential to improve spectral efficiency and network performance [9]. Nonetheless, provisioning of reliable communication for both PU and SU nodes is a challenging task, as PU nodes have preemptive priority over SU nodes for spectrum access; wherefore the key issue is to ensure that SU nodes do not degrade performance of PU nodes. In general, DSA can be categorized into: overlay, underlay and hybrid modes [1], [14]. In overlay, SU nodes select free PU channels and instantly refrain from further transmissions on detecting PU arrival on any of the occupied channels. In underlay, SU nodes transmit through channels that are occupied by PU nodes, but at lower power levels for which the resulting co-channel interference is below tolerable limits as seen by PU nodes [14]-[15]. Hybrid mode merges the merits of overlay and underlay modes by adopting both for SU transmissions.

Transmit-power control Spectrum-management Radio environment Radio-scene analysis Spectrum holes Noise-floor statistics Traffic statistics Interference temperature Quantized channel capacity Action: transmitted signal RF Stimuli Channel-state estimation Predictive modeling

Figure 1-1: Basic cognitive radio functions and operational cycle [13].

(23)

1.1.2 Radio Resource Management

Radio resource management (RRM) techniques mainly deal with maximizing radio resource utilization and efficiency in wireless networks, while guaranteeing the varying QoS requirements for different users. In the case of CRNs, the issue of spectrum agility complicates RRM, making it even more challenging. Information from both the PN and CRN is required for efficient spectrum allocation in the CRN, as shown in Fig. 1-2. Thus, efficient spectrum management and resource allocation are some of the cornerstones on which performance of SU nodes can be enhanced, while protecting PU nodes at the same time. The commonly employed methods for resource allocation and optimization techniques are based on (i) heuristic methods, (ii) game theory, (iii) graph theory, (iv) convex optimization and (v) stochastic modelling to mention a few [1], [7].

1.1.2.1 Heuristic Methods

Heuristic based resource allocation techniques in wireless networks provide relatively easy solutions that can be computed within acceptable time and complexity [7]. Although heuristic methods are mostly useful when dealing with optimization problems that are very hard to solve, they usually provide suboptimal solutions without any guarantee for convergence and optimality. Furthermore, heuristic methods have generally been found to be more suitable for obtaining quick solutions when employed CRNs, whereby resource allocation problems are otherwise mainly characterized by challenging optimization problems with high complexity [16].

CRN Resource Optimization Secondary Network Information Primary Network Information

Figure 1-2: Resource allocation and optimization process in CRNs.

User Selection Channel Allocation

(24)

CHABALALA S.C. 4 2017 Ph.D. THESIS 1.1.2.2 Game Theory

In general, game theory provides a mathematical framework to model interaction of multiple entities (i.e. players) in a competition (i.e. game), whose decisions affect one another [3], [17]. The players aim to maximize their utility on each decision taken. Game theory can be categorized into cooperative and non-cooperative models. In cooperative models, players collaborate to exchange information and make decisions that improve the overall network utility, in which case the solution point is commonly known as Nash bargaining [18]. In the case of non-cooperative models, players make selfish decisions to maximize their individual utility irrespective of how their decisions affect other players, and the common solution point is referred to as Nash equilibrium [19].

1.1.2.3 Graph Theory

Graph theory is one of the techniques that have been employed to solve scheduling and resource optimization problems in wireless networks, whereby a graph (X, Y) represents an optimization model with vertices X representing network entities, and the edges Y for their interactions. In particular, there are various types of graph modelling for resource allocation problems, such as bipartite graph, vertex coloring and conflict graph, as discussed in detail in [20]. Moreover, the types of graphs can further be classified as directed (i.e. whereby the edges have directions) or undirected, and weighted (i.e. whereby each edge is assigned a nonnegative weight) or unweighted [21]. The graph based models have mainly been employed in infrastructure based networks, with a central controller for gathering network information and performing resource allocation.

1.1.2.4 Machine Learning Algorithms

Machine learning concepts such as neural networks, fuzzy logic and genetic algorithms are not uncommon for RRM and spectrum allocation in CRNs. Neural networks have been found to be more suitable for highly dynamic environments with frequent radio spectrum changes, to which SU nodes are required to promptly respond [22]. Fuzzy logic is based on human understandable fuzzy sets and inference rules to obtain resource allocation solutions. Fuzzy logic based techniques are also relatively simple as they are not based on complicated mathematical models [23]-[24]. These are also appropriate for real-time CRN applications with stringent constraints on system response time. Genetic algorithms are based on evolutionary biological processes to determine optimal solutions for complex problems, but with fast convergence and ease of implementation [25].

(25)

1.1.2.5 Convex Optimization

Resource allocation problems are formulated as optimization models with objective functions for which an optimal solution can be obtained among all the feasible solutions. In the case of CRNs, resource allocation problems are usually formulated as constrained optimization problems that take resource constraints into consideration [1]-[3], [7], [9], [11]. The optimization problems can further be classified into linear, non-linear, convex, non-convex integer or mixed-integer non-linear programming (MINLP) problems [26]. Different optimization tools can be employed to determine optimal solutions based on the structure and nature of the optimization problems. For example, Lagrangian framework with dual decomposition can be employed to solve convex optimization problems, while linear programming techniques can be employed to solve linear optimization problems where the objective function and the associated constraints are all linear [27].

1.1.3 Channel Assembling in Wireless Networks

Channel assembling (ChA) constitutes one of the emerging spectrum access techniques with a great potential to support the envisaged heterogeneity in future wireless networks. This is a technique through which a user performs concurrent transmissions through multiple channels to maximize capacity and improve spectral efficiency [27]-[29]. In general, ChA techniques can be classified into static and dynamic schemes. A predetermined fixed number of channels is employed in the case of static schemes, whereas the number of assembled channels varies in the case of dynamic ChA schemes; hence, the number of assembled channels can be adjusted based on QoS requirements for different applications and network constraints. In this thesis, ChA refers to both spectrum bonding and aggregation, wherefore the arrangement of assembled channels with respect to one another in frequency domain is insignificant. In a strict sense however, a set of contiguous non-overlapping channels is required for channel bonding, as illustrated in Fig. 1-3; while channel aggregation does not necessarily require the assembled channels to be contiguous [29]-[30].

channel-aggregation channel-bonding

Figure 1-3: Illustration of channel-bonding and channel-aggregation.

Frequency P o w er Z[\ ]^_ ]^` ]^a d^e b[e d^c b[c

(26)

Notwithstanding the disregard of the arrangement of assembled channels in this thesis, it has been established in the previous studies that channel aggregation is relatively subject to higher complexity and overhead than channel bonding, as a result of the need for channel management and scheduling policies across aggregated channels [29], [31]-[32]. In the case of CRNs, PU nodes have exclusive rights to access radio spectrum. Thus, increasing the number of assembled channels for SU nodes increases the probability that a PU node may appear on any of the selected channels, which therefore increases collision probability between PU services and SU services [27]-[28], [31], [33]. In general, maximizing capacity for SU services through ChA schemes is a challenging task that conflicts with the issue of spectrum agility with respect to the PU nodes.

1.2

Problem Statement and Research Objectives

1.2.1 Problem Statement

The ubiquity of wireless devices and the proliferation of network applications with stringent QoS requirements are in contrast to the issue of the limited radio spectrum, in which case the conventional fixed spectrum management and allocation techniques have been found to be a bottleneck for high spectral efficiency and utilization by the Federal Communications Commission (FCC) [3]. The concept of CR is continuously evolving, but fulfilling QoS requirements for SU nodes in CRNs remains is a critical concern, that is also extremely challenging due to spectrum agility whereby preemptive priority is granted to licensed users. The applicability of CRNs in addressing the increasing demands for high speed wireless networks therefore requires meticulous and insightful design considerations for development of efficient RRM techniques and DRA schemes. Efficient RRM techniques are imperative for reliable and robust communication.

The central issue in this thesis is therefore development of efficient resource allocation and ChA schemes for CRNs. In particular, this focuses on how to perform adaptive resource allocation and ChA taking into account PU activity patterns, fading wireless channels, interference constraints to protect PU transmissions, QoS requirements for SU services, network dynamics and resource constraints. This compels development of appropriate spectrum characterization techniques for selection of channels that offer optimal performance for CRNs, as well as reducing susceptibility to performance degradation due to PU activity patterns and network dynamics.

(27)

CHABALALA S.C. 7 2017 Ph.D. THESIS 1.2.2 Primary Objectives

The primary aims and objectives of the research presented in this thesis are:

(a) To study and identify the challenges associated with efficient RRM in CRNs, and investigate the intrinsic factors that affect performance of ChA schemes and RA techniques.

(b) To develop adaptive ChA and RA schemes that determine the optimal number of channels and power distribution for overlay CRNs, subject to power constraint, QoS requirement on capacity, PU activities and collision probability threshold over fading channels.

(c) To develop statistical characterization of assembled channels in fading channels, and derive compact closed-form analytical models to quantify and gauge performance of ChA and RA schemes on average capacity, outage probability and forced termination probability.

(d) To develop criteria on how to perform adaptive ChA and RA for hybrid overlay and underlay CRNs architectures, with respect to PU spectrum occupancy and interference constraints imposed on SU transmissions to protect ongoing PU services.

(e) To develop adaptive and dynamic ChA and RA schemes for multiuser multichannel access in CRNs, with the aim to maximize total network capacity, subject to time varying wireless channels, individual users QoS requirements, PU activities and resource constraints.

In summary, this research is mainly focused on the development of new techniques for adaptive ChA and RA schemes in CRNs, with the aim to improve performance and provide reassurance for satisfying QoS requirements for SU services, without degrading performance of PU nodes.

1.3

Original Contributions and Thesis Organization

Unlike the presented work in this thesis, the main research studies on RA techniques based on the Lagrangian framework do not incorporate PU activity patterns as a constraint in formulating optimization problems. In the case of ChA schemes, the key studies in literature are mainly based on continuous time Markov chain (CTMC) modeling with a predetermined and fixed number of channels to assemble. Moreover, the issue of transmit power optimization has been generally

(28)

overlooked in CTMC based studies. In this regard, the main contributions of this work are derived from the aforementioned primary aims and objectives, accordingly outlined as follows:

(a) A criterion for determining the optimal channel selection, together with the associated power distribution for ChA schemes in overlay CRNs has been developed. This takes into account, the time varying and fading nature of wireless channels, PU nodes activity patterns, QoS requirements on capacity for SU transmissions, and the constraint on SU nodes total transmission power to make judicious ChA decisions.

(b) The statistical characterization of assembled channels in terms of probability distribution function (PDF) and cumulative distribution function (CDF) has been derived. This characterization provides the foundation on which closed-form mathematical models are derived towards theoretical performance analysis and evaluation of ChA schemes.

(c) Closed-form analytical models have been developed to quantify and evaluate performance of ChA schemes in terms of the average capacity, outage probability and forced termination probability for SU transmissions. Further, the correctness and preciseness of the analytical models have been confirmed through cross-validation with extensive simulations.

(d) Adaptive hybrid overlay and underlay ChA scheme that aims to maximize SU capacity in spectrum sharing wireless networks has been developed. The developed scheme determines the optimal number of channels to assemble in both overlay mode and underlay for SU transmissions. Multilevel maximum allowable SU transmit power in underlay mode has been derived to ensure that SU transmit power is constrained under the noise floor of PU nodes. (e) As an extension to the optimal ChA technique mentioned in (a) above, optimal ChA scheme

for multiuser multichannel access in overlay CRNs has been developed. The optimal scheme is based on convex optimization and Lagrangian relaxation framework. Alternatively, a suboptimal ChA scheme has been developed based on the modified Hungarian algorithm. Although the developed ChA and RA schemes may be complex in terms of practical implementation issues, the work in this thesis presents in-depth analysis and evaluation which provide an exposition for significant theoretical insights into the performance gains that can be obtained by employing optimization techniques in CRNs and future wireless networks in general.

(29)

1.4

Author Publications

The following are peer-reviewed journals and conference publications:

• C.S. Chabalala and F. Takawira, “Hybrid channel assembling and power allocation for multichannel spectrum sharing wireless networks,” accepted for the IEEE Wireless Communications and Networking Conference (WCNC’2017), San Francisco, CA, March 2017.

• C.S. Chabalala, R. van Olst and F. Takawira, “Optimal channel selection and power allocation for assembling in cognitive radio networks,” in Proceedings of the IEEE Global Communications Conference (GLOBECOM’2015), San Diego, CA, pp. 1-6, December 2015.

• C.S. Chabalala and F. Takawira, “Performance analysis of channels selection and power allocation for channel assembling in multichannel cognitive radio networks,” in review for publication in IEEE Transactions on Vehicular Technology, submitted November 2016.

• C.S. Chabalala and F. Takawira, “Hybrid overlay and underlay spectrum aggregation with optimal channel selection and adaptive power allocation in cognitive radio networks,” submitted for publication in the IET Communications Journal, submitted February 2016.

• C.S. Chabalala and F. Takawira, “Adaptive spectrum aggregation for opportunistic resource allocation in multichannel wireless networks,” to be submitted for the Proceedings of the IEEE AFRICON’2017, Cape-Town, 18-20 September, 2017.

1.5

Thesis Organization

The following provides the roadmap for the remainder of this thesis:

Chapter 2: Presents the overview of the investigation into the related works in literature, mainly focusing on the DSA techniques for CRNs, RRM and optimization techniques in wireless communications, and ChA schemes with more emphasis on CRNs.

Chapter 3: Presents the adaptive ChA and RA scheme which determines the optimal number of channels to assemble, as well as the associated optimal power profile for SU transmissions in overlay CRNs. This chapter also provides the details of the employed Lagrangian relaxation

(30)

framework that is based on convex optimization theory, together with the branch-and-bound (BnB) technique with sequential fixing (SF) for optimal solution to the original MINLP problem. Moreover, simulation based performance evaluation is presented for various ChA schemes.

Chapter 4: This chapter presents performance modelling of ChA schemes over fading channels. The closed-from expressions for the PDF and CDF are derived for statistical characterization of assembled channels. The characterization forms the basis on which the compact closed-form mathematical models for average capacity, outage probability and forced-termination probability are derived. Then the preciseness and correctness of the derived analytical models are established by performing cross-validation between simulation results and analytical results, wherefore the cross-validation is employed to confirm the accuracy of the developed analytical models.

Chapter 5: This chapter presents adaptive ChA schemes for hybrid overlay and underlay SU transmissions. This allows SU nodes to opportunistically assemble free PU channels in overlay mode, and adaptively select the occupied PU channels with controlled transmit power levels in underlay mode. A multilevel maximum allowable SU transmit power for channels assembled in underlay mode is derived based on various assumptions about the knowledge of channel status information (CSI) availability at the SU transmitter, which ensures that SU transmit power in underlay mode is constrained under the noise floor of PU nodes. Further, the Lambert-L function is employed to derive closed-form expressions for channel selection. Then finally, the simulation based performance evaluation for hybrid ChA schemes is presented.

Chapter 6: In this chapter, the adaptive ChA schemes for multiuser multichannel access CRNs are presented. In particular, two ChA schemes are discussed. First, an optimal scheme that is based on convex optimization to determine the overall optimal solution for multiple SU nodes is presented, followed by a suboptimal ChA scheme that employs the modified Hungarian algorithm. Then, the simulation based performance analysis is presented to evaluate performance of the ChA schemes for multiuser multichannel access CRNs, subject to PU activity patterns.

Chapter 7: This chapter provides the main points on which the work presented in this thesis is summarised, together with the concluding remarks. This also highlights the possible directions on how the accomplished work can be extended for future research on adaptive ChA schemes and efficient RA techniques for CRNs and the emerging wireless networks.

(31)

C

HAPTER

T

WO

2

L

ITERATURE

R

EVIEW

2.1

Introduction

HIS chapter presents a comprehensive review of the related works on recent advances on ChA and RA techniques in literature. The previous research trends and accomplishments are therefore summarised with more emphasis on CRNs, wherein the susceptibility to performance degradation due to PU activity patterns and network dynamics compel development of robust ChA and RA techniques to improve performance of SU nodes. Since the emergence of CR technology, development of appropriate communication techniques for efficient operation of CRNs has always been a challenge due to radio spectrum dynamics as dictated by the activities of PU nodes. Thus, the presented review provides the core pillars and foundations which serve as a preamble to the research ideas and contributions that are discussed in subsequent chapters.

The rest of the chapter is organized as follows: Section 2.2 presents the related works on DSA techniques in CRNs, where in particular, the overlay mode, underlay mode and hybrid access mode are discussed. Section 2.3 delves into research accomplish on ChA schemes, underlining the open issues and challenges in the context of CRNs. Then Section 2.4 presents the related works on RA techniques in wireless communications and spectrum sharing networks, as well as highlighting the optimization frameworks that are usually employed; followed by Section 2.5 which summarizes the main points to conclude the work presented in this chapter.

2.2

Dynamic Spectrum Access in Cognitive Radio Networks

One of the major challenges in CRNs is the provisioning of efficient and robust communication without degrading performance of PU nodes. Robust communication refers to the ability of a network to operate continuously under changing environments and resource constraints without failure [1], [3], [7]. To address the issue of robust communication amid the looming overcrowding of radio spectrum, the subsections in sequel provide DSA techniques and research efforts in CRNs.

T

(32)

CHABALALA S.C. 12 2017 Ph.D. THESIS 2.2.1 Overlay Spectrum Access

In overlay spectrum access mode, SU nodes opportunistically select the free PU channels, and instantly vacate the occupied channels on detection of new arrivals for PU services. This is usually referred to as opportunistic spectrum access (OSA) [13]. The vacation of SU nodes as a result of PU arrivals results in forced terminations, which therefore degrades performance of SU nodes. The overlay spectrum access forms the basis of operation for the original inception of CR technology. SU nodes are therefore strictly required to detect spectrum holes (white spaces) defined in space, time and frequency as illustrated in Fig. 2-1. This also requires accurate online spectrum sensing techniques to promptly detect arbitrary arrivals of new PU services. However, the previous works in literature have shown that accurate spectrum sensing is difficult to implement [1], [34].

The main objective in OSA is collision avoidance between SU and PU services, which is analogous to interference mitigation whereby PU nodes have exclusive rights to spectrum access. One of the constraints on which SU nodes can make spectrum access decisions is the collision probability [27]-[28], [31]. This is the probability that a PU service arrives on a channel that is occupied by SU nodes. To optimize performance in OSA CRNs, joint spectrum sensing and access problem has been formulated as a partially observable Markov decision process (POMDP) in [35]; whereby PU activities have been modeled as a two-state Markov chain process (i.e. idle or busy). The key focus was to maximize throughput while maintaining collision probability below a threshold. However, the state space of the POMDP model increases with the number of channels; hence, increase in complexity, based on which a suboptimal scheme was developed to reduce complexity [35].

Time Power

Figure 2-1: The concept of spectrum holes in CRNs.

Spectrum holes (opportunities)

(33)

In [34], the authors proposed OSA scheme whereby reactive PU nodes take the history of SU activity patterns into consideration to determine the probability that a particular channel can be accessed. The PU channel occupancy is modeled as a 4-state discrete Markov chain process. The main focus of the study was to maximize SU throughput by formulating the optimization problem as a constrained finite-horizon POMDP. Then the numerical results were presented to illustrate that the proposed scheme guaranteed satisfactory QoS level on throughput for PU nodes, while making a tradeoff to increase spectrum access opportunities for SU transmissions in a CRN. Similar works that are based on POMDP have also been reported in [36]-[38], where in general, POMDP based problems were found to be computationally prohibitive to solve.

In addition to sensing outcomes about channel occupancy status, the works in [27], [31], [39]-[43] proposed robust spectrum access policies that incorporate the time varying quality of wireless channels. The key objective in these works is to maximize effective throughput of SU nodes subject to channel variations and collision constraints. In general, it was established that capturing the quality of wireless channels for spectrum access improves performance under tights collision constraints. However, in the case where PU activity patterns are high, spectrum is hardly available for SU nodes while operating in overlay mode, to the extent that SU nodes would completely have no access to spectrum when PU nodes are always busy [27], [31]. Development of appropriate DSA access techniques that can cope with high PU activities is therefore required.

2.2.2 Underlay Spectrum Access

In contrast to the overlay spectrum access, SU nodes in underlay mode employ interference threshold to mitigate performance degradation in high PU activity patterns [7], [44]-[45]. This facilitates concurrent spectrum access for both PU and SU transmissions; where SU nodes transmit through the channels that are occupied by PU nodes, subject to transmit power and interference control as illustrated on Fig. 2-2 on the next page. Thus, minimizing interference for SU transmissions is the most common criterion for efficient communication. In particular, interference constraint is imposed on SU transmissions to protect PU services. Nonetheless, co-channel interference is usually bidirectional, therefore affects both PU and SU transmissions alike. How to maintain interference threshold for SU transmissions while guaranteeing QoS requirements for both PU services and SU services is therefore a technically challenging task [2], [7].

(34)

Various studies on underlay CRNs have incorporated PU outage probability as a basis for interference management and power control. In most studies, interference constraints and transmit power constraints are usually employed to protect PU nodes from SU transmissions. Interference constraint refers to the maximum amount of interference that a PU node can tolerate without QoS degradation [1], [3], [15]. In [46]-[50], peak interference constraint has been employed to analyse performance and improve capacity of SU nodes in underlay CRNs over fading channels, where in particular, adaptive power allocation for SU transmissions has been mainly based on the SNR at the PU receiver. The results in these studies revealed significant capacity gain for CRNs over fading channels, where full knowledge of CSI is available at the SU transmitter.

Other studies in [51]-[55] investigated the impact of imperfect knowledge of CSI at the SU transmitter, whereby a closed-form expression for SU capacity was derived based on peak power constraint. Moreover, transmission power allocation techniques have been proposed in [56] to minimize outage probability in Rayleigh fading wireless channels. From the previous studies, it has been established that using outage constraint at the PU node receiver offers better performance and protection for PU services than using interference power constraint [56].

As noted in [56], full CSI at the SU transmitter is required to protect PU transmissions using interference power constraint; yet full CSI may be difficult to obtain as full cooperation between PU nodes and SU nodes is required. Outage probability constraint may be relatively easier to implement as it rather based on statistical information [57]. The impact of PU transmissions on SU nodes was also investigated under Rayleigh fading channels in [58]-[59], where it was revealed that PU transmissions can also result in severe fading against SU transmissions.

P o w er Frequency Interference threshold

Figure 2-2: Example of underlay spectrum access in CRNs.

PU transmission

(35)

2.2.3 Hybrid Overlay and Underlay Spectrum Access

Hybrid spectrum access mode mainly aims to merge the merits of the overlay and underlay modes by adopting both for SU transmissions as they have their respective advantages [3], [15], [60]-[61]. Hence, hybrid schemes jointly exploit the overlay and the underlay spectrum access techniques as illustrated in Fig. 2-3. In previous studies, it has been shown that hybrid schemes outperform either overlay-only or underlay-only schemes in terms of achievable system capacity and bit-error-rate (BER) [14]-[15], [62]-[65]. In [66]-[67], hybrid overlay and underlay schemes were studied for CRNs over additive white Gaussian noise (AWGN) channels, where it was generally established that hybrid schemes achieve significant performance improvement in CRNs.

Spectrum sensing forms the critical component for transmit mode selection and switching in hybrid schemes [8], [61], [68]. The study in [63] proposed a Markov chain model that facilitates the switching between overlay and underlay spectrum access modes. The proposed model detects PU activity patterns based on a double-threshold energy detection technique to mitigate collision and interference between PU and SU services. In [14], a hybrid overlay and underlay spectrum access was investigated, with the aim to maximize data-rate subject to power constraint; whereby an auction-based power allocation technique was employed for competing SU nodes.

Moreover, a location-aware spectrum access scheme was proposed in [8], whereby SU nodes which are close to a PU node access spectrum in overlay mode, while concurrent spectrum access in underlay mode is allowed for SU nodes that are located far from a PU node. It was generally established that incorporating location information in spectrum access techniques improves spectrum efficiency. However, it was also found that the location-aware approach mainly depends

P o w er Frequency Interference threshold

Figure 2-3: Example of hybrid spectrum access in CRNs.

PU node

(36)

on network topology, especially in terms of the distance between SU transmitter in a CRN, and a PU receiver in a PN [3], [7]-[8]. The study in [70] proposed a hybrid scheme that aims to improve throughput and spectrum efficiency, as well as reducing congestion in CRNs. SU services with minimum flow are allowed to bypass sensing phase to access spectrum under hybrid mode subject to signal-to-noise and interference ratio (SNIR) constraints; while unoccupied channels are left for SU services with higher flows. The authors presented simulation results, which revealed improved performance and spectrum utilization. Another study in [71] presented a power control scheme based on game theory, where SU nodes compete for spectrum access. A repeated game model was used to ensure that SNIR at the PU receiver is kept below a predefined threshold [70]-[72].

2.3

Channel Assembling in Wireless Communications

This section presents the related works on ChA schemes in literature. In recent works, channel ChA technique has been proposed and implemented in practical networks, as a mechanism to maximize capacity in wireless communications [27]-[28], [33], [73]-[75]. This is a technique through which a user performs concurrent transmissions through multiple channels. Arguably, ChA is the only effective solution for network applications with large amounts of data to send, but through narrow frequency bands [76]. In literature, static and dynamic ChA techniques have been proposed [28], [31], [73]-[78]. For static ChA schemes, the number of assembled channels is prefixed, while the number of channels may vary for dynamic ChA schemes. In an effort to establish the performance benefits and limitations associated with ChA schemes, experimental and theoretical studies have been reported in [27]-[33], [64], [73]-[85], where it was revealed that naïve ChA decisions can greatly degrade performance in wireless networks, as the benefits of ChA are greatly influenced by various network factors such as noise levels and interference from neighboring links.

The key research studies on ChA in CRNs have mainly concentrated on CTMC modeling to provide significant insights into performance of ChA schemes [27]. Most of the reported studies are thus, predominantly focused on medium access control (MAC) based performance analyses. In [80], the authors proposed a spectrum sensing scheme which employs channel bonding and maintains a list of backup channels for redundancy, with the aim to meet various QoS requirements for heterogeneous SU services while reducing spectrum access latency. A similar study has also been reported in [81] for construction of a network backbone aiming to improve network reliability

(37)

and throughput. In accord with the study [80], it has been demonstrated in [81] that channel bonding with backup-list maintenance provides significant performance improvement. In [27]-[28], [73]-[75], dynamic ChA where SU nodes adjust the number of channels based on heterogeneous traffic and channel availability during transmissions has been presented. In general, increasing the number of assembled channels for SU nodes increases capacity, hence reduces transmission time and latency. Reducing transmission time reduces probability of collisions between SU nodes and PU nodes. However, increasing the number of channels decreases SNR per channel, and also increases susceptibility to collision with PU service arrivals. Thus, determining the optimal number of channels subject to collision constraints in CRNs is also a challenging task [27].

A recent study on the performance of ChA in underlay CRNs has been presented in [86]. Similar to majority of the existing studies in literature, the work in [86] is mainly based on CTMC modeling in a multiuser environment. Further, comprehensive studies based on CTMC analytical framework to investigate performance of channel bonding under various network conditions have been presented in [31], [76], [79], [82]. In these studies, factors that affect performance, and conditions under which channel bonding schemes improve performance in wireless networks were established, whereby it has been generally concluded that significant performance benefits of ChA schemes can be achieved by adaptively adjusting channel-width with respect to QoS requirements and network conditions among other parameters. However, the CTMC based analyses in literature do not necessarily account for the issues pertaining to network dynamics and constraints such as the varying nature of wireless channels and adaptive power control. Moreover, the study in [83] proposed the channel aggregation diversity (CAD) scheme with joint channel selection and power allocation, which aims to improve spectrum efficiency and energy efficiency bounded by total power constraint. Through simulations studies, the CAD scheme has been found to offer improved performance. Unlike the work presented in this thesis, a node using CAD scheme selects a set of channels with predetermined power levels through the exchange of control packets.

Further, ChA scheme based on queuing and channel fragmentation was proposed in [77], and another with priority based queuing in [78]. Thus, queuing and fragmentation were found to reduce blocking probability of SU nodes significantly. However, the results are also limited to MAC schemes without appropriate spectrum characterization for efficient ChA. Optimizing channel selection and power distribution for ChA schemes in CRNs requires spectrum assessment and

(38)

characterization to determine the quality of available channels. In general, stronger signal strength is usually required for transmission through assembled channels to achieve the same reliability of a single channel [30]-[31], which therefore mandates judicious design considerations for development of efficient ChA schemes. For instance, doubling the number of channels decreases signal strength by 3dB [32]. This is because the same amount of transmit power is distributed across all the channels; wherefore the power allocated per channel is reduced. Moreover, allocation of several channels to a single SU node through ChA schemes reduces the probability of channel availability for arrival of new SU services, hence may have detrimental effects on the performance of SU nodes in a CRN as a result of increased blocking probability [27], [30, [31]-[32], [79]. In general, the major efforts in prior works mainly focused on a preset number of assembled channels, with the same power levels across the selected channels, independent of the number of channels assembled. Hence, the total power increases linearly with the number of channels. However, results obtained in such manner are misleading if applied to SU nodes with a total power constraint as illustrated in Fig 2-4. Also, impact of the fading nature of wireless channels have been generally overlooked, yet ChA schemes are inherently prone to fading as it results in performance degradation. Thus, how to optimize channel selection and power distribution for ChA schemes is a

−10 −5 0 5 10 15 20 25 30 0 2 4 6 8 10 12 14 16 18x 10 7 SNR (dB)

Channel Capacity (bits/sec)

1x6MHz − Single Channel 2x6MHz − Constant Power 2x6MHz − Shared Power 3x6MHz − Constant Power 3x6MHz − Shared Power

(39)

crucial issue that needs to be investigated. Accordingly, the optimization compels development of appropriate spectrum assessment and characterization criteria for available channels. To augment the highlighted limitations, this thesis extends the studies in literature, with the main objective to develop criteria on how to perform adaptive ChA with the optimal number of channels and power profile. This entails selection of the best channels to assemble, together with the optimal power distribution to maximize SU capacity, taking into account: channel variations, total power constraint, minimum capacity threshold, and collision probability threshold to protect PU services. This aims to reassure QoS requirements satisfaction while minimizing outage probability under fading channels, reduce collision probability between PU and SU services, hence reduce forced termination probability for SU services. The departure of SU nodes as a result of PU service arrivals results in forced terminations, which generally degrades performance of SU nodes.

2.4

Dynamic Resource Allocation in Wireless Networks

This section highlights the related works on optimization techniques for dynamic resource allocation (DRA) in wireless networks, which also includes CRNs. Different techniques for development of optimal and suboptimal solutions have been studied in previous works. For instance, greedy and heuristic resource allocation schemes have been proposed, whereby the greedy schemes have been found to be effective in homogenous environments where all the users require the same amount of spectrum [87]; in which case, the amount of spectrum allocated to users is predetermined and fixed, whereas adapting the allocated spectrum based on QoS requirements has been found to be more effective and efficient; especially in spectrum sharing networks where the spectrum allocation schemes also incorporate PU traffic patterns [2], [4], [8]-[11], [87].

Moreover, various studies have looked into development of mathematical models to establish performance insights of resource allocation techniques based on: game theory for problem formulation, CTMC modelling and convex optimization [1], [2]-[3]. Many of the studies based on CTMC modelling in spectrum sharing networks mainly overlooked the effects of the varying nature of wireless channels [27], [64]. As a result, these studies do not account for the issues related to adaptive RRM and network dynamics such as PU activity patterns in the case of spectrum sharing wireless networks, where adaptive resource allocation schemes based on PU activity patterns are otherwise crucial to improve performance of both PUs and SUs. Nonetheless, the dependence of the

(40)

performance of RRM schemes on wireless channel links and transmit power optimization highlights the necessity for development of appropriate schemes that are resilient to performance degradation [3], [7], [88]-[89]. One of the popular approaches on DRA techniques in wireless communications is convex optimization technique, whereby convex reformulation is performed by relaxing the binary integer constraints such as _Q,f∈ #0,1%, which usually denotes channel selection status for user i on channel j. Convex reformulation is mainly performed by the following relaxation k0 ≤ _Q,f≤ 1m, which is normally interpreted as sharing factor [26]. In general, constrained optimization problems are usually expressed as follows [1], [87], [90]:

minimize_∈r s( ) (2.1)

subject to: } ~ ( ) ≤ 0, for = 1, ⋯ , j

ℎƒ( ) = 0, for „ = 1, ⋯ , i,

where s( ) is the objective function for the optimization problem, ~ ( ) and ℎ_ƒ( ) are respectively inequality and equality constraints, and is the optimization parameter vector with the feasible range r. In most studies, convex optimization framework is employed with Lagrangian decomposition, whereby the Lagrangian reformulation for (2.1) can be expressed as

ℒ( , †, ‡) = s( ) − ‰ 5 ~ ( ) f Š‹ − ‰ Œƒℎƒ( ) Q ƒ (2.2)

where ∀5 ∈ † and ∀Œ_ƒ∈ ‡ are Lagrangian multipliers [26], [41], [91]-[94]. This is usually solved by determining the Karush-Kuhn-Tucker (KKT) conditions which are necessary and sufficient for optimality; based on which closed-form expressions can be developed, which are however, difficult to obtain in most cases [26]-[27]. It has been shown in previous works such as in [95]-[97] that most near-optimal heuristic approaches that solve (2.1) would have a complexity order of Ž(ijS). The Lagrangian relaxation with dual optimization framework on the other hand, achieves

99.9999% of the optimal solution with complexity order of Ž(ij); hence the optimality gaps are

usually less than 10•‘ with Lagrangian duality theorem [97]. In orthogonal frequency division multiplexing (OFDM) systems, convex optimization with dual decomposition has been mostly employed for joint subcarrier allocation and power distribution in multiuser systems [95]-[99].

(41)

One of the popular issues on DRA techniques in wireless networks is fairness, which indicates how equally resources are allocated in multiuser systems [8], [10], [100]. For instance, fairness may refer to equal allocation of channels to multiple users, or allocation of equal portions of transmit power based on the total power budget, or rate proportionality which aims to achieve the same rate for each user in a network [101]-[103]. In terms of rate proportionality, max-min techniques have been employed in various studies in literature, whereby the main objective is to maximize the minimum data rate of users [103]. The rate proportionality based methods have also been employed for efficient and fair distribution of resources in heterogeneous environments [100], [104].

In the case of CRNs, DRA and optimization techniques are more challenging than in conventional wireless networks as a result of the imperative necessity to protect PU nodes against SU transmissions. Most of the works on resource allocation in CRNs in general, overlap with the studies discussed in Section 2.2. Some of the previous works studied rate and power allocation problems in multichannel access CRNs, where the optimization problems considered rate and power allocation schemes to satisfy QoS requirements for SU services, and power control to minimize interference imposed on PU services from SU transmissions [2], [7]-[9], [41], [55]-[56]. Performance of SU nodes is absolutely dictated by PU activity patterns in CRNs, which are however, mostly overlooked in the existing works on resource allocation techniques.

Notwithstanding, QoS-aware resource allocation in CRNs is a challenging task due to network dynamics and time varying wireless channels, as well as the obligation to provide efficient communication for SU services without degrading performance of PU nodes. Further, it is worth reiterating once again that, the main studies on resource allocation techniques in spectrum sharing networks are mainly focused on user selection to m

Channel assembling and resource allocation in multichannel spectrum sharing wireless networks

NIVERSITY OF THE

UTHORIZATION

DECLARATION

EDICATION

ACKNOWLEDGEMENTS

ABSTRACT

TABLE OF CONTENTS

List of Figures... xii

CHAPTER 2 Literature Review ... 11

CHAPTER 4 Performance Modelling and Evaluation ... 47

CHAPTER 6 Multiuser Multichannel Spectrum Access ... 86

CHAPTER 7 Conclusions and Future Works ... 103

Appendix B ... 111

LIST OF FIGURES

LIST OF TABLES

LIST OF ACRONYMS

LIST OF SYMBOLS

HAPTER

Figure 1-1: Basic cognitive radio functions and operational cycle [13].

Figure 1-2: Resource allocation and optimization process in CRNs.

1.1.3 Channel Assembling in Wireless Networks

Problem Statement and Research Objectives

1.2.1 Problem Statement

Original Contributions and Thesis Organization

Author Publications

Thesis Organization

HAPTER

Dynamic Spectrum Access in Cognitive Radio Networks

Figure 2-1: The concept of spectrum holes in CRNs.

2.2.2 Underlay Spectrum Access

Figure 2-2: Example of underlay spectrum access in CRNs.

Figure 2-3: Example of hybrid spectrum access in CRNs.

Channel Assembling in Wireless Communications

Dynamic Resource Allocation in Wireless Networks